DeepAir: Predicting Radio Spectrum Usage at Scale with Deep Temporal Convolutional Networks

The rapid uptake of wireless technologies over the past decade has resulted in an increasing pressure on the limited radio spectrum resources. To improve the efficiency of the current spectrum allocation, dynamic spectrum sharing is being considered by regulators in several jurisdictions. The success, however, of an efficient dynamic wireless environment depends on the ability to characterize spectrum usage patterns. Since traditional methods prove unable to scale to a wide range of channels, we propose DeepAir, a robust and scalable model that is capable of learning and predicting complex temporal and spectral patterns in multivariate spectrum data. Specifically, we design a Sequence-to-Sequence model that employs an encoder-decoder architecture with two Deep Temporal Convolutional Networks.Using a test set consisting of approximately 900 channels in the Land Mobile Radio frequency bands, we obtain a median RMSE and median MAE of approximately 6.51 and 5.15, respectively, for a one-week forecast horizon. Moreover, we apply transfer learning to demonstrate the effectiveness of this model in forecasting patterns from any sensor, regardless of the band, the sensitivity, and the geographical location. Finally, our proposed approach demonstrates no significant performance degradation when predicting usage almost three years after training.